Preprints
https://doi.org/10.5194/egusphere-2026-3537
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/egusphere-2026-3537
© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
10 Jul 2026
| 10 Jul 2026
Status: this preprint is open for discussion and under review for Biogeosciences (BG).
Ideas and perspectives: Addressing environmental challenges using distributed data generation: From collaborative networks to artificial intelligence-enabled science
Abstract. Distributed data generation, or data collected from multiple sources and locations using standardized approaches and involving coordination among investigators, has emerged as a powerful approach to meet contemporary demands for scalable environmental knowledge. However, practitioners often lack guidance on best practices for distributed data generation, and a framework classifying its modalities is missing.
To address these gaps, we developed a conceptual framework organizing distributed data generation along two axes: participant-based (ranging from highly formalized to highly flexible) and method-based (from experimental to observational). This framework provides common vocabulary across modalities and describes how different approaches affect data generation logistics and outcomes. We propose operational best practices across three critical pillars: outreach, operations, and output (i.e., publications, data), leveraging lessons learned from over 35 existing distributed data projects. Lastly, we explore how emerging artificial intelligence (AI) capabilities may help address longstanding challenges in distributed data generation, including in coordination, adaptive sampling, and cross-project data integration. This perspective provides strategies and identifies opportunities to advance distributed data generation for addressing pressing biogeochemical, environmental, and societal challenges. We underscore the transformative potential of distributed data generation for modern, broad-scale environmental research, and provide guidance on how to realize that potential.
How to cite. Goldman, A. E., Stegen, J. C., Borer, E. T., Hensley, R., Myers-Pigg, A., Borton, M. A., Muradyan, P., Ward, N. D., Andino, J. M., Richardson, A. D., Leverkus, A. B., Lowry, C. S., McDowell, N. G., Patel, K. F., Davidson, S. J., Forbes, B., Downs, M. R., Bolton, W. R., Scheibe, T. D., and Malhotra, A.: Ideas and perspectives: Addressing environmental challenges using distributed data generation: From collaborative networks to artificial intelligence-enabled science, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2026-3537, 2026.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
Download & links
Download & links
- Preprint
(733 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 23 Sep 2026)
Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor
| : Report abuse
- RC1: 'Comment on egusphere-2026-3537', Christoph Praschl, 07 Aug 2026 reply
Viewed
Total article views: 272 (including HTML, PDF, and XML)
Cumulative views and downloads
(calculated since 10 Jul 2026)
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 200 | 60 | 12 | 272 | 25 | 18 |
- HTML: 200
- PDF: 60
- XML: 12
- Total: 272
- BibTeX: 25
- EndNote: 18
Viewed (geographical distribution)
Total article views: 125 (including HTML, PDF, and XML)
Thereof 125 with geography defined
and 0 with unknown origin.
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
1
Latest update: 23 Aug 2026
Energy and Environment Directorate, Pacific Northwest National Laboratory, Richland, WA, 99352, USA
James C. Stegen
Earth and Biological Sciences Directorate, Pacific Northwest National Laboratory, Richland, WA 99352, USA
School of the Environment, Washington State University, Pullman, WA, 99163, USA
Elizabeth T. Borer
Department of Ecology, Evolution, and Behavior, University of Minnesota, Saint Paul, MN, 55108, USA
Robert Hensley
National Ecological Observatory Network, Battelle, Boulder, CO, 80301, USA
Allison Myers-Pigg
Earth and Biological Sciences Directorate, Pacific Northwest National Laboratory, Sequim, WA, 98382, USA
Mikayla A. Borton
Department of Food Science and Human Nutrition, Colorado State University, Fort Collins, CO, 80526, USA
Paytsar Muradyan
Environmental Science Division, Argonne National Laboratory, Lemont, IL, 60439, USA
Nicholas D. Ward
Energy and Environment Directorate, Pacific Northwest National Laboratory, Sequim, WA, 98382, USA
Jean M. Andino
Arizona State University, School for the Engineering of Matter, Transport and Energy, Tempe, AZ, 85226, USA
Andrew D. Richardson
School of Informatics, Computing, and Cyber Systems, Northern Arizona University, Flagstaff, AZ, 86011, USA
Center for Ecosystem Science and Society, Northern Arizona University, Flagstaff, AZ, 86011, USA
Alexandro B. Leverkus
Departamento de Ecología, Facultad de Ciencias, Universidad de Granada, Av. Fuentenueva s/n, 18071, Spain
Christopher S. Lowry
University at Buffalo, Department of Earth Sciences, Buffalo, NY, 14260, USA
Nate G. McDowell
Earth and Biological Sciences Directorate, Pacific Northwest National Laboratory, Richland, WA 99352, USA
School of the Environment, Washington State University, Pullman, WA, 99163, USA
Kaizad F. Patel
Earth and Biological Sciences Directorate, Pacific Northwest National Laboratory, Richland, WA 99352, USA
Scott J. Davidson
Département des sciences biologiques, Université du Québec à Montréal, Montréal, QC, H2X 1Y4, Canada
Brieanne Forbes
Energy and Environment Directorate, Pacific Northwest National Laboratory, Richland, WA, 99352, USA
Martha R. Downs
Long-Term Ecological Research Network, National Center for Ecological Analysis and Synthesis, University of California Santa Barbara, Santa Barbara, CA, 93101, USA
W. Robert Bolton
Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, TN, 37830, USA
Timothy D. Scheibe
Earth and Biological Sciences Directorate, Pacific Northwest National Laboratory, Richland, WA 99352, USA
Earth and Biological Sciences Directorate, Pacific Northwest National Laboratory, Richland, WA 99352, USA
Short summary
Distributed data generation, or data collected from multiple sources and locations using standardized approaches and coordination among researchers, is powerful for scalable environmental knowledge. This paper helps readers understand options for distributed data generation, offers practical guidance for outreach, operations, and sharing outcomes, and explores how artificial intelligence can help address longstanding challenges.
Distributed data generation, or data collected from multiple sources and locations using...
The authors synthesize experience from a 2025 workshop (25+ practitioners) and an assessment of 35+ distributed data efforts into three contributions: (1) a conceptual framework that organizes distributed data generation along two continua - a participant-based axis (highly formalized <-> highly flexible) and a method-based axis (experimental <-> observational); (2) a catalogue of best practices grouped under three pillars - outreach, operations, and output - anchored by an extensive example table (Table 1); and (3) a forward-looking discussion of how AI may address longstanding challenges in co-design, adaptive sampling, and data harmonization.
The paper is well written, timely, and addresses a real gap: the lack of shared vocabulary and cross-domain guidance for distributed environmental science. The two-axis framework (Sect. 2) is the most durable and original contribution and is likely to be very useful to the community. The best-practices synthesis (Sect. 3, Table 1) is a valuable resource.
MAJOR COMMENTS
1. Section 4.1 is bound to a specific technology (LLMs) rather than to the function it serves - and is too narrowly text-centric
This is my central concern. Section 4.1 is framed as "LLM-driven enhancements..." and the entire argument is built on large language models as THE mechanism for co-design and cross-domain translation. This is a problem for a Perspective paper, which should age well, on two counts:
(a) Technology-binding. LLMs are the dominant technology today, but the field is moving fast and the specific technology may change within the lifetime of this paper. The underlying capability the authors actually need is an interface: technical or domain-specific content goes in, and content optimized for a different audience comes out (e.g., soil scientist <-> modeler <-> farmer <-> policy-maker). That capability is defined by its function - content in, audience-appropriate content out - not by whichever model happens to implement it. It could be an LLM now; it could be a different architecture, a hybrid symbolic/neural system, or something not yet named later. I strongly recommend the authors reframe Section 4.1 (and its heading) around the function/interface rather than the named technology, and treat "LLMs" as the current-best instantiation cited as an example, not as the organizing concept. This would future-proof the section and make its argument more robust.
(b) Text-centrism / missing modalities. The section treats co-design and science-to-action almost entirely as a text problem ("translate plans, findings, and implications," "convert highly technical content into content... to convert to policy"). But effective communication and co-design across disparate audiences - farmers, land managers, rights holders, the public, policy-makers - very often depends on non-text modalities: infographics, data visualizations, maps, diagrams, decision-support dashboards, and audio/visual material. These require different underlying technologies (generative imagery, automated data visualization, cartographic tools), not just language models. Given that the paper elsewhere rightly emphasizes multi-modal engagement (Sect. 3, AmeriFlux example) and reducing barriers for non-technical participants, it is inconsistent for the AI-outreach section to collapse back to a purely text-in/text-out view. I recommend the authors explicitly broaden Section 4.1 to acknowledge that communication is multi-modal and that realizing this vision will require a portfolio of technologies serving the same co-design function.
(c) Internal inconsistency of "altitude" across Section 4. Reinforcing the point above: the three subsection headings mix levels of abstraction. Section 4.2 is framed neutrally as "AI-guided..." (a function/capability), whereas 4.1 and 4.3 lead with the specific technology ("LLM-driven...", and LLM-centric text in 4.3). I suggest the authors adopt a consistent framing throughout Section 4: name each subsection by the challenge or function being addressed (translation/co-design, adaptive sampling, harmonization), and cite specific technologies (LLMs, agents, etc.) as current examples within. This single change would resolve both the technology-binding concern and the inconsistency.
2. Justify the collapse of standardization and centralization into a single participant axis
Section 2.1 builds on SanClements et al. (2022), collapsing "degree of standardization" and "degree of centralization" into one axis while reframing it around participants (formalized <-> flexible). The manuscript notes SanClements et al. proposed a 1:1 relationship between the two, but this 1:1 assumption is not obviously robust: one can imagine efforts that are highly centralized in funding/management yet loosely standardized in protocols, or vice versa. Because this collapse is foundational to the framework, please add a sentence or two justifying it and acknowledging cases where standardization and centralization decouple (and where a project might therefore sit ambiguously on the axis). Relatedly, "formalized/flexible" (engagement structure) and "standardized/unstandardized" (protocols) are related but not identical concepts; a brief clarification of exactly what the axis measures would help readers place their own efforts.
3. Scope statement vs. examples (human-on-the-ground coordination)
Section 1 explicitly scopes the paper to "efforts with human-to-human coordination requiring on-the-ground efforts." However, several featured examples are largely autonomous/sensor-based - most notably Argo (Table 1), a global array of autonomous profiling floats, and to a degree the flux-tower and streamgaging networks. Please reconcile the scoping statement with these examples: either broaden the scope statement, or clarify that the human-coordination criterion applies to the network's design and governance rather than to each measurement. As written there is a tension a careful reader will notice.
4. Evidence base and its own geographic/institutional bias
The paper rightly flags that existing networks are heavily biased toward western Europe and North America (Sect. 3). It is worth noting explicitly that the manuscript's own evidence base - a 2025 workshop and the 35+ exemplar projects, many of them author-affiliated and US-DOE/PNNL-centric (WHONDRS, EXCHANGE, MONet, NGEE Arctic, CROCUS, GROWdb) - is subject to the same selection bias. This does not undermine the synthesis, but a sentence acknowledging the provenance and potential representativeness limits of the exemplar set would improve transparency and match the self-critical tone the authors apply elsewhere.
5. Missing: CARE principles alongside FAIR
The output pillar and conclusions lean heavily on FAIR (Wilkinson et al., 2016), which is appropriate. But the paper repeatedly invokes co-design with "rights holders," communities, and the public, and the equitable-participation literature. In that context, the CARE Principles for Indigenous Data Governance (Collective benefit, Authority to control, Responsibility, Ethics) are now a standard complement to FAIR for community-generated and Indigenous data and would strengthen both Section 3 (output) and the equity framing. Their absence is a notable gap given the paper's emphasis on broad, equitable engagement.
MINOR / SPECIFIC COMMENTS
- Citation mismatch (Sect. 4.1, p. 13, ~l. 291): The sentence "expediting cross-domain translation, similar to LLM use in cross-language translation (Gui et al., 2019)" is supported by Gui et al. (2019), Geoforum - "Globalization of science and international scientific collaboration: A network perspective." That paper is about collaboration-network geography and predates modern LLMs; it does not support a claim about LLM-based (machine) cross-language translation. Or do I oversee the point? Please replace with an appropriate reference on neural/LLM machine translation, or reword.
- Citation mismatch (Sect. 2.2.1, ~l. 186): "DroughtNet (Poyatos et al., 2021)." The cited reference (Poyatos, Werner & Martinez-Vilalta, 2021) is the SAPFLUXNET sap-flow database, not DroughtNet (the International Drought Experiment network). Please correct the citation or the network name.
- Consistency of the core term. The abstract defines distributed data generation as "data collected from multiple sources and locations," while the Introduction uses "geographically dispersed locations." Minor, but please make the definition verbatim consistent between abstract and main text.
- Section 3 reads partly as an annotated bibliography. The best-practices section is valuable but at times enumerates examples more than it distills transferable principles. Consider foregrounding the principle for each best practice (one crisp sentence) and relegating the project instances to Table 1, so the takeaways are easier to extract.
- The authors already note "concrete examples of AI-enabled distributed efforts remain limited." It may help readers to state explicitly that Sections 2-3 (framework and best practices) are the durable contribution while Section 4 is deliberately speculative/horizon-scanning - this manages expectations and protects the paper's longevity.
TECHNICAL CORRECTIONS
- Fig. 1 caption: "using multi-model engagement" appears to be a typo for "multi-modal engagement" (the term used consistently in Sect. 3).
- Sect. 4.2 (~l. 347): "(U.S. DOE 2026b." - missing closing parenthesis.